REVIEWS

38 Turner, T.F. et al. (2000) Nested cladistic analysis indicates population

fragmentation shapes genetic diversity in a freshwater mussel.
Genetics 154, 777785
39 Birks, H.J.B. (1989) Holocene isochrone maps and patterns of treespreading in the British Isles. J. Biogeog. 16, 503540
40 Sinclair, W.T. et al. (1999) The postglacial history of Scots pine (Pinus
sylvestris L.) in western Europe: evidence from mitochondrial DNA
variation. Mol. Ecol. 8, 8388
41 Willis, K.J. and Whittaker, R.J. (2000) Paleoecology - The refugial
debate. Science 287, 14061407
42 Stauffer, C. et al. (1999) Phylogeography and postglacial colonization
routes of Ips typographus L. (Coleoptera, Scolytidae). Mol. Ecol. 8, 763773
43 Holder, K. et al. (1999) A test of the glacial refugium hypothesis using
patterns of mitochondrial and nuclear DNA sequence variation in rock
ptarmigan (Lagopus mutus). Evolution 53, 19361950
44 Hewitt, G.M. (1993) After the ice: Parallelus meets Erythropus in the

45
46

47

48

49

Pyrenees. In Hybrid Zones and the Evolutionary Process (Harrison,

R.G., ed), pp. 140164, Oxford University Press
Whitlock, M.C. and McCauley, D.E. (1999) Indirect measures of gene
flow and migration: FST <> 1/(4Nm + 1). Heredity 82, 117125
Moritz, C. and Faith, D.P. (1998) Comparative phylogeography and the
identification of genetically divergent areas for conservation. Mol.
Ecol. 7, 419429
Templeton, A.R. and Georgiadis, N.J. (1996) A landscape approach to
conservation genetics: conserving evolutionary processes in the
African Bovidae. In Conservation Genetics: Case Histories from Nature
(Avise, J.C. and Hamrick, J.L., eds), pp. 398430, Chapman & Hall
Templeton, A.R. et al. (1992) A cladistic analysis of phenotypic associations
and haplotypes inferred from restriction endonuclease mapping and
sequence data. III. Cladogram estimation. Genetics 132, 619633
Posada, D. et al. (2000) A program for the cladistic nested analysis of
the geographical distribution of genetic haplotypes. Mol. Ecol. 9, 487

Statistical methods for detecting

molecular adaptation
Ziheng Yang and Joseph P. Bielawski

t has been proved remarkably difficult to get compelling evidence for changes
in enzymes brought about by
selection, not to speak of adaptive
changes1.

The past few years have seen the

development of powerful statistical
methods for detecting adaptive molecular
evolution. These methods compare
synonymous and nonsynonymous
substitution rates in protein-coding genes,
and regard a nonsynonymous rate
elevated above the synonymous rate as
evidence for darwinian selection.
Numerous cases of molecular adaptation
are being identified in various systems
from viruses to humans. Although
previous analyses averaging rates over
sites and time have little power, recent
methods designed to detect positive
selection at individual sites and lineages
have been successful. Here, we summarize
recent statistical methods for detecting
molecular adaptation, and discuss their
limitations and possible improvements.

and weaknesses , so that they can

be used to detect more cases of
molecular adaptation.

Measuring selection using the

nonsynonymous/synonymous
(dN/dS) rate ratio
Although Darwins theory of
Traditionally, synonymous and
evolution by natural selection is
nonsynonymous
substitution
generally accepted by biologists
rates (Box 1) are defined in the
for morphological traits (includcontext of comparing two DNA
ing behavioural and physiologisequences, with dS and dN as the
cal), the importance of natural
numbers of synonymous and nonselection in molecular evolution
synonymous substitutions per
has long been a matter of debate.
site, respectively5. Thus, the ratio
The neutral theory2 maintains
v 5 dN/dS measures the difference
that most observed molecular
between the two rates and is most
variation both polymorphism
easily understood from a mathewithin species and divergence
matical description of a codon
between species is due to ransubstitution model (Box 2). If an
dom fixation of selectively neutral
amino acid change is neutral, it
mutations. Well established cases
will be fixed at the same rate as a
of molecular adaptation have Ziheng Yang and Joseph Bielawski are at the Galton synonymous mutation, with v 5
Laboratory, Dept of Biology, University College
been rare3. Several tests of neu1. If the amino acid change is delLondon, 4 Stephenson Way, London, UK NW1 2HE
trality have been developed and
eterious, purifying selection (Box
(z.yang@ucl.ac.uk; j.bielawski@ucl.ac.uk).
1) will reduce its fixation rate,
applied to real data, and although
they are powerful enough to
thus v , 1. Only when the amino
reject strict neutrality in many
acid change offers a selective
genes, they rarely provide unadvantage is it fixed at a higher
equivocal evidence for positive darwinian selection.
rate than a synonymous mutation, with v . 1. Therefore,
Most convincing cases of adaptive molecular evolution an v ratio significantly higher than one is convincing
have been identified through comparison of synonymous evidence for diversifying selection.
(silent; dS) and nonsynonymous (amino acid-changing; dN)
The codon-based analysis (Box 2) cannot infer whether
substitution rates in protein-coding DNA sequences, thus synonymous substitutions are driven by mutation or selecproviding fascinating case studies of natural selection in tion, but it does not assume that synonymous substitutions
action on the protein molecule. Selected examples are are neutral. For example, highly biased codon usage can be
listed in Table 1; see Hughes4 for detailed descriptions of caused by both mutational bias and selection (e.g. for transmany case studies. Here, we summarize recent method- lational efficiency6 ), and can greatly affect synonymous
ological developments that improve the power to detect substitution rates. However, by employing parameters pj for
adaptive molecular evolution, and examine their strengths the frequency of codon j in the model (Box 2), estimation of

496

0169-5347/00/$ see front matter 2000 Elsevier Science Ltd. All rights reserved.

PII: S0169-5347(00)01994-7

TREE vol. 15, no. 12 December 2000

REVIEWS

Table 1. Selected examples of protein-coding genes in which positive selection was detected by using the dN/dS ratio
Gene

Organism

Refs

Genes involved in defensive systems or immunity

Class I chitinase gene

Colicin genes
Defensin genes
Fv1
Immunoglobulin VH genes
MHC genes
Polygalacturonase inhibitor
genes
RH blood group and RH50
genes
Ribonuclease genes
Transferrin gene
Type I interferon-v gene
a1-Proteinase inhibitor genes

41
45
46
47
48
49
50

Primates and rodents

51

Primates
Salmonid fishes
Mammals
Rodents

52
53
54
55

Envelope gene
gH glycoprotein gene
Hemagglutinin gene
Invasion plasmid antigen genes
Merozoite surface antigen-1 gene
msp 1a
nef
Outer membrane protein gene
Polygalacturonase genes
Porin protein 1 gene
S and HE glycoprotein genes
Sigma-1 protein gene
Virulence determinant gene

FMD virus
Plasmodium falciparum
Hepatitis D virus
Phages G4, fX174, and
S13
HIV
Pseudorabies virus
Human influenza A virus
Shigella
Plasmodium falciparum
Anaplasma marginale
HIV
Chlamydia
Fungal pathogens
Neisseria
Murine coronavirus
Reovirus
Yersinia

substitution rates will fully account for codon-usage bias

(Box 1), irrespective of its source. Because parameter v is a
measure of selective pressure on a protein, it differentiates
codon-based analyses from the more general tests of
neutrality proposed in population genetics7,8. These general
tests often lack the power to determine the sources of the
departure from the strict neutral model, such as changes in
population size, fluctuating environment or different forms
of selection.

Estimation of dN and dS between two sequences

Two classes of methods have been suggested to estimate
dN and dS between two protein-coding DNA sequences. The
first class includes over a dozen intuitive methods developed since the early 1980s (Refs 5,915). These methods
involve the following steps: counting synonymous (S) and
nonsynonymous (N) sites in the two sequences, counting
synonymous and nonsynonymous differences between the
two sequences, and correcting for multiple substitutions
at the same site. S and N are defined as the sequence length
multiplied by the proportions of synonymous and nonsynonymous changes before selection on the protein14,16. Most
of these methods make simplistic assumptions about the
nucleotide substitution process and also involve ad hoc
treatment of the data that cannot be justified14,15; therefore, we refer to these methods of estimating dN and dS
as approximate methods. The methods of Miyata and
Yasunaga5, and Nei and Gojobori9, assume an equal rate for
TREE vol. 15, no. 12 December 2000

Organism

Refs

Genes involved in reproduction

Arabis and Arabidopsis

Escherichia coli
Rodents
Mus
Mammals
Mammals
Legume and dicots

Genes involved in evading defensive systems or immunity

Capsid gene
CSP, TRAP, MSA-2 and PF83
Delta-antigen coding region
E gene

Gene

42
56
57
3
40
3
33
3
58
3
38
3
50
59
60
3
3

18-kDa fertilization protein gene

Acp26Aa
Androgen-binding protein gene
Bindin gene
Egg-laying hormone genes
Ods homeobox gene
Pem homeobox gene
Protamine P1 gene
Sperm lysin gene
S-Rnase gene
Sry gene

Abalone (Haliotis)
Drosophila
Rodents
Echinometra
Aplysia californica
Drosophila
Rodents
Primates
Abalone (Haliotis)
Rosaceae
Primates

61
62
63
64
3
65
66
67
61
68
69

Bovids
Primates

70
23

Conus gastropods
Crotalinae snakes

71
72

Genes involved in digestion

k-casein gene
Lysozyme gene
Toxin protein genes

Conotoxin genes
Phospholipase A2 gene

Genes related to electron transport and/or ATP synthesis

ATP synthase F0 subunit gene

COX7A isoform genes
COX4 gene

Escherichia coli
Primates
Primates

3
73
74

Rodents
Primates
Rodents

75
75
75

Saccharomyces cerevisiae
Vertebrates
Antarctic fishes
Drosophila
Rat

3
76
77
78
3

Cytokine genes

Granulocyte-macrophage SF gene
Interleukin-3 gene
Interleukin-4 gene
Miscellaneous

CDC6
Growth hormone gene
Hemoglobin b -chain gene
Jingwei
Prostatein peptide C3 gene

transitions (T C and A G) and transversions (T,C

A,G), as well as a uniform codon usage. Because transitions
at the third wobble position are more likely to be
synonymous than transversions, ignoring the transition/
transversion rate ratio leads to underestimation of S and
overestimation of N (Ref. 10). Efforts have been taken to
incorporate the transition/transversion rate bias (Box 1)
when counting sites and differences1014. The effect of
Box 1. Glossary
Codon-usage bias: unequal codon frequencies in a gene.
Nonsynonymous substitution: a nucleotide substitution that changes the
encoded amino acid.
Prior probability: the probability of an event (such as a site belonging to a
site class) before the collection of data.
Positive selection: darwinian selection fixing advantageous mutations with
positive selective coefficients. The term is used interchangeably with molecular adaptation and adaptive molecular evolution.
Posterior probability: the probability of an event conditional on the
observed data, which reflects both the prior assumption and information in
the data.
Purifying selection: natural selection against deleterious mutations with
negative selective coefficients. The term is used interchangeably with negative selection or selective constraints.
Synonymous substitution: a nucleotide substitution that does not change
the encoded amino acid.
Transition/transversion rate bias: unequal substitution rates between
nucleotides, with a higher rate for transitions (changes between T and C and
between A and G) than transversions (all other changes).

497

REVIEWS

Box 2. A model of codon substitution

The codon is considered the unit of evolution. The substitution rate from
codons i to j (i j) is given as:
0,

j ,

qij = j ,

j ,
,
j

if i and j differ at more than one position,

for synonymous transversion,
for synonymous transition,
for nonsynonymous transversion,
for nonsynonymous transition.

Parameter k is the transition/transversion rate ratio, pj is the equilibrium frequency of codon j and v (5 dN /dS) measures the selective pressure on the
protein. The qij are relative rates because time and rate are confounded in
such an analysis. Given the rate matrix Q 5 {qij}, the transition probability
matrix over time t is calculated as:
P(t) = {pij(t)} = eQt

where pij(t) is the probability that codon i becomes codon j after time t. Likelihood calculation on a phylogeny involves summing over all possible codons
in extinct ancestors (internal nodes of the tree). After Refs 16,18,27,79.

biased codon usage has largely been ignored17; however,

extreme codon-usage bias can have devastating effects on
the estimation of dN and dS (see the next section)15,18. A
recent ad hoc method15 incorporates both transition and
codon-usage biases.
The second class is the maximum likelihood (ML)
method based on explicit models of codon substitution
(Box 2)16,19. Parameters in the model (i.e. sequence
divergence t, transition/transversion rate ratio k and the
dN/dS ratio v) are estimated from the data by ML, and are
used to calculate dN and dS according to their definitions15,16,20. A major feature of the method is that the
model is formulated at the level of instantaneous rates
(where there is no possibility for multiple changes) and
that probability theory accomplishes all difficult tasks in
one step: estimating mutational parameters, such as k;
correcting for multiple hits; and weighting pathways of
change between codons.

Statistical tests can be used to test whether dN is significantly higher than dS. For approximate methods, a normal
approximation is applied to dN 2 dS. For ML, a likelihoodratio test can be used. In this case, the null model has v
fixed at 1, whereas the alternative model estimates v as a
free parameter. Twice the log-likelihood difference
between the two models is compared with a x2
distribution with one degree of freedom to test whether v
is different from one.
Computer simulation has been used to examine the
performance of different estimation methods; the findings
are consistent with observations made in real data analyses14,15,19. We demonstrate the effects of different estimation procedures using human and orangutan a2-globin
genes (Table 2). For comparison, different assumptions are
made in ML concerning the transition/transversion rate
bias and the codon-usage bias. The simpler models are
each rejected when compared with more complex models
by likelihood-ratio tests, confirming biased transition rates
and codon usage. Thus, estimates from ML accounting for
both biases (Model 8, Table 2) are expected to be the most
reliable. We make the following observations:
Assumptions appear to matter more than methods.
The approximate methods and ML produce similar
results under similar assumptions. The method of
Nei and Gojobori is similar to ML under a model
that ignores both transition/transversion bias and
codon-usage bias (Model 1, Table 2), whereas the
methods of Ina and Li are similar to ML under a
model accounting for the transition/transversion
bias but ignoring codon-usage bias (Model 2, Table
2). The method of Yang and Nielsen15 is similar to
ML under a model accounting for both biases
(Model 6, Table 2). However, for distantly related
sequences, ad hoc treatment in approximate methods can lead to serious biases even under the correct assumptions19.
Ignoring the transition/transversion rate bias
leads to underestimation of S, overestimation of
dS and underestimation of the v ratio10.
Codon-usage bias in these data has the opposite

Table 2. Estimation of dN and dS between the human and orangutan a2-globin genes (142 codons)a
Method and/or model

dN

dS

dN /dS (v)

,c

Refs

1.0

2.1
6.1

109.4
NA
119.3
61.7

316.6
NA
299.9
367.3

0.0095
0.0104
0.0101
0.0083

0.0569
0.0517
0.0523
0.1065

0.168
0.201
0.193
0.078

9
11
14
15

1.0
3.0
1.0
3.9
1.0
5.4
1.0
5.3

108.5
124.6
129.1
137.1
63.2
60.6
58.3
55.3

317.5
301.4
296.9
288.9
362.8
365.4
367.7
370.7

0.0093
0.0099
0.0092
0.0093
0.0084
0.0084
0.0082
0.0082

0.0557
0.0480
0.0671
0.0624
0.0973
0.1061
0.1145
0.1237

0.167
0.206
0.137
0.149
0.087
0.079
0.072
0.066

2633.67
2632.47
2612.40
2610.48
2560.76
2557.85
2501.39
2498.61

16
16
16
16
16
16
16
16

Approximate methods

Nei and Gojobori

Li
Ina
Yang and Nielsen
ML methodsb

(1) Fequal, k 5 1
(2) Fequal, k estimated
(3) F134, k 5 1 fixed
(4) F134, k estimated
(5) F334, k 5 1 fixed
(6) F334, k estimated
(7) F61, k 5 1 fixed
(8) F61, k estimated
aGenBank

accession numbers are V00516 (human) and M12158 (orangutan).

equal codon frequencies (5 1/61) are assumed; F134, four nucleotide frequencies are used to calculate codon frequencies (3 free parameters);
F334, nucleotide frequencies at three codon positions are used to calculate codon frequencies (9 free parameters); F61, all codon frequencies are used as
free parameters (60 free parameters).
c, is the log-likelihood value.
bFequal,

498

TREE vol. 15, no. 12 December 2000

REVIEWS
effect to the transition/transversion bias; ignoring
codon-usage bias leads to overestimation of S,
underestimation of dS and overestimation of v.
This gene is extremely GC-rich at the third codon
position, with base frequencies at 9% (T), 52% (C),
1% (A) and 37% (G). Most changes at the third
position (before selection at the amino acid level)
are transversions between C and G. Thus, the
number of synonymous sites is less than half that
expected under equal base and codon frequencies. Although, in theory, the bias caused by
unequal codon frequencies can be in the opposite
direction15, we have not encountered a real gene
showing that pattern. Such codon-usage bias
appears to have misled previous analyses examining the relationship between the GC content at
silent sites and dS, because those studies ignored
the codon-usage bias when estimating dS (Ref. 21).
Different methods can produce different estimates, even when the sequences are highly similar. The sequences used in Table 2 are only about
10% different at silent sites and ,1% different at
nonsynonymous sites; however, estimates of v
are three times different. Because all estimation
procedures partition the total numbers of sites
and differences into synonymous and nonsynonymous categories, underestimation of one means
overestimation of the other, thus resulting in large
errors in the v ratio.

Detecting lineage-specific episodes of

darwinian selection
If, for most of the time, a gene evolves under purifying
selection but is occasionally subject to episodes of adaptive change22, a comparison between two distantly related
sequences is unlikely to yield a dN/dS ratio significantly
greater than one. Methods have been developed to detect
positive selection (Box 1) along specific lineages on a phylogeny. If the gene sequences of the extinct ancestors were
known, it would be straightforward to use the pairwise
methods discussed above. Thus, Messier and Stewart23
inferred ancestral lysozyme gene sequences through phylogenetic analysis24,25, and used them to calculate dN and dS
for each branch in the phylogeny. Their analysis identified
two lineages in a primate phylogeny with highly elevated
nonsynonymous substitution rates. The same approach
was taken in a test of relaxed selective constraint in the
rhodopsin gene of cave-dwelling crayfishes26.
There are also likelihood models that allow different v
ratios for branches in a phylogeny18,27. Using such models,
likelihood-ratio tests can be constructed to test hypotheses.
For example, the v ratio for a predefined lineage can be
either fixed at one or estimated as a free parameter. The likelihood values under those two models can be compared, to
test whether v . 1 in that lineage. Similarly, a model assuming a single v for all lineages (the one-ratio model) can be
compared with another model assuming an independent v
for each lineage (the free-ratio model), to test the neutral
prediction that the v ratio is identical among lineages18,27.
It should be noted that variation in the v ratio among
lineages is a violation of the strictly neutral model2,18,28,29, but
it is not sufficient evidence for adaptive evolution. In particular, if nonsynonymous mutations are slightly deleterious,
they will have a higher probability of fixation in a small population than in a large one30, and thus lineages of different
population sizes will have different v ratios. Besides positive
selection, relaxed selective constraint can also elevate the v
TREE vol. 15, no. 12 December 2000

Box 3. Likelihood and Bayes

The statistical-estimation theory used in the methods discussed in this
review can be explained with the following simple hypothetical example.
Suppose that a population is an admixture of two groups of people in the
proportions 60% and 40%, and a certain disease occurs at a rate of 1% in
Group I and of 0.1% in Group II. Suppose a random sample of 100 individuals
is taken from the population, what is the probability that three of them carry
the disease? The probability that a random individual carries the disease (D)
is an average over the two groups (G1 and G2):
p = P (D) = P (G1) P (DG1) + P (G2) P (DG2) = 0.6 0.01 + 0.4 0.001 = 0.0064 (1)

Similarly, the probability that an individual does not carry the disease is:

P (D ) = P (G1) P (DG1) + P (G2) P (DG2)

= 0.6 0.99 + 0.4 0.999 = 0.9936 = 1 p

(2)

The probability that three out of 100 individuals carry the disease is given by
the binomial probability:

P=

100! 3
p (1 p )97 = 0.0227

3!x97!

(3)

If Eqn 3 involves an unknown parameter [such as the rate P (D|G1) in Group I],
that parameter can be estimated by maximizing Eqn 3. In that case, Eqn 3
gives the probability of observing the data (sample) and is called the likelihood function.
The second question is to calculate the probability that an individual in the
sample who carries the disease is from Group I. The Bayes theorem gives
this probability as:
P (G1D) = P (G1) P (DG1)/P (D) = 0.6 0.01/0.0064 = 0.94

(4)

Note that this is just the proportion of the contribution from Group I to P(D) in
Eqn 1. Thus, this individual is most likely to be from Group I. Similarly, a
healthy individual in the sample is more likely to be from Group I than from
Group II because

P (G1D ) = P (G1) P (D G1)/P (D ) = 0.6 0.99/0.9936 = 0.5978

and P (G2D ) = 1 P (G1D ) = 0.4022

(5)
In methods for inferring sites under positive selection36,37, we let D in the
example be the data at a site and Gi be the ith site class with the dN/dS ratio
v i. The probability of observing data at a site is then an average over the site
classes (Eqn 1). The product of such probabilities over sites constitutes the
likelihood (Eqn 3), from which we estimate any unknown parameters, such
as the branch lengths and parameters in the v distribution over sites. After
the parameters are estimated, we use the Bayes theorem to calculate the
probability that any site, given data at that site, is from each site class (Eqns
4 and 5).
Another straightforward application of the theory is ancestral sequence
reconstruction; in this case, we replace Gi with a reconstruction (characters
at interior nodes of the phylogeny) at a site. When we calculate the likelihood
function, the probability of data at a site P (D) is a sum over all possible ancestral reconstructions (G i s) (Eqns 1 and 2). After parameters are estimated, the
reconstruction that makes the greatest contribution to P (D) is the most likely
(Eqns 4 and 5)24.
The Bayes method discussed here is known as the empirical Bayes, because
it uses estimates of parameters and does not account for their sam-pling errors.
This might be a concern if parameters are estimated from small samples or if the
posterior probabilities are sensitive to parameter estimates. An alternative
approach is the hierarchical Bayes method, which accounts for the uncertainty in
unknown parameters by averaging over their prior distribution.
Note that the reconstructed ancestral sequences24, as well as the inferred
site classes in the site-class models36,37, are pseudo data and involve systematic biases. To appreciate such biases, note that in the previous example,
the Bayes calculations (Eqns 4 and 5) predict that each of the 100 individuals
in the sample, healthy or sick, are from Group I. Although this is the best
prediction, the accuracy is low. If such inferred group identities are used for
further statistical analysis, misleading results might follow.

ratio it might be difficult to distinguish the two if the

estimated v is not larger than one. Furthermore, it is incorrect
to use the free-ratio model to identify lineages of interest and
then to perform further tests on the v ratios for those
lineages using the same data without any correction27.

499

REVIEWS

Posterior probability

(a)
0
0.2
0.4
0.6
0.8
1.0
RSWHYVEPKFLNKAFEVALKVQIIAGFDRGLVKWLRVHGRTLSTVQKKALYFVNRRYMQTHWANYMLWINKKIDALGRTPVVGDYTRLGAEIGRRIDMAYFYDFLKDKNMIPKYLPYMEEINRMRPADVPVKYM

Sites in lysin

(b)
74

13

82

123

95

99
116
107
38

44

Fig. 1. The identification of sites under positive selection from the sperm lysin genes of 25
abalone species. (a) Posterior probabilities for site classes with different v ratios along the
sequence. The lysin sequence of the red abalone (Haliotis rufescens) is shown below the x-axis.
ML estimates under Model M3 (discrete)37 suggest three site classes with the v ratios at
v0 5 0.085 (grey), v1 5 0.911 (green) and v2 5 3.065 (red), and with proportions p0 5 0.329,
p1 5 0.402 and p2 5 0.269. These proportions are the prior probabilities (Box 1) that any site
belongs to the three classes. The data (codon configurations in different species) at a site alter
the prior probabilities dramatically, and thus the posterior probabilities might be different from
the prior probabilities. For example, the posterior probabilities for Site 1 are 0.944, 0.056 and
0.000, and thus this site is likely to be under strong purifying selection. The posterior probabilities
for Site 4 are 0.000, 0.000 and 1.000, and thus this site is almost certainly under diversifying
selection. (b) Lysin crystal structure from the red abalone (Protein Data Bank file 1LIS), with sites
coloured according to their most likely class inferred in (a). The structure starts from amino acid
four (His) at the N-terminus, because the first three amino acids are unresolved. The five ahelices are indicated: a1 from amino acids 13 to 38, a2 from 44 to 74, a3 from 82 to 95, a4 from
99 to 107 and a5 from 116 to 123. Note that sites potentially under positive selection (red) are
scattered all over the primary sequence but tend to cluster around the top and bottom of the crystal structure. Reproduced, with permission, from Ref. 39.

Methods based on ancestral reconstruction might not

provide reliable statistical tests because they ignore errors
and biases in reconstructed ancestral sequences (Box 3).
The ML method has the advantage of not relying on reconstructed ancestral sequences. It can also easily incorporate features of DNA sequence evolution, such as the transition/transversion rate bias and codon-usage bias, and is
thus based on a more realistic evolutionary model. When
likelihood-ratio tests suggest adaptive evolution along certain lineages, ancestral reconstruction might be useful to
pinpoint the involved amino acids and to infer ancestral
proteins, which can be synthesized and examined in the
laboratory31,32.

Detecting amino acid sites under darwinian selection

The methods discussed so far assume that all amino acid
sites are under the same selective pressure, with the same
v ratio. The analysis effectively averages the v ratio across
all sites and positive selection is detected only if that average is .1. This appears to be a conservative test of positive
selection because many sites might be under strong purifying selection owing to functional constraint, with the v
ratio close to zero.
A few recent studies addressed this problem. Fitch and
colleagues33,34 used parsimony to reconstruct ancestral
DNA sequences, and counted changes at each codon site
along branches of the tree. They tested whether the
proportion of nonsynonymous substitutions at each site
is greater than the average over all sites in the sequence.
Suzuki and Gojobori35 took a more systematic approach.
For each site in the sequence, they estimated the
numbers of synonymous and nonsynonymous sites and
differences along the tree using reconstructed ancestral
sequences, and then tested whether the proportion of

500

Trends in Ecology & Evolution

nonsynonymous substitutions differed from the neutral

expectation (v 5 1). Suzuki and Gojoboris criterion is
more stringent than Fitch et al.s, because the v ratio averaged over sites is almost always , 1. These methods are
expected to require many sequences in the data set so that
there are enough changes at individual sites. Furthermore,
the reliability of significance values produced by these
methods might be affected by the use of ancestral reconstruction, which is most unreliable at the positively
selected or variable sites24, and by codon composition
bias, which is most extreme at a single site.
In a likelihood model, it is impractical to use one v parameter for each site. The standard approach is to use a
statistical distribution to describe the variation of v among
sites; for example, we might assume several classes of sites
in the protein with different v ratios36,37. The test of positive selection then involves two major steps: first, to test
whether sites exist where v . 1, which is achieved by a
likelihood-ratio test comparing a model that does not allow
for such sites with a more general model that does; and
second, to use the Bayes theorem to identify positively
selected sites when they exist. Sites having high posterior
probabilities (Box 1) for site classes with v . 1 are potential targets of diversifying selection. The theory is
explained in Box 3 (Refs 20,36,37).
Nielsen and Yang36 implemented a likelihood-ratio test
based on two simple models. The null model, M1 (neutral),
assumes a class of conserved sites with v 5 0 and another
class of neutral sites with v 5 1. The alternative model, M2
(selection), adds a third class of sites with v estimated
from the data. (The model codes are those used in the
codeml program in the PAML package.) If M2 fits the data
significantly better than M1 and the estimated v ratio for
the third class in M2 is .1, then some sites are under
TREE vol. 15, no. 12 December 2000

REVIEWS
diversifying selection. Zanotto et al.38 used this test to identify several sites under strong positive selection in the nef
gene of HIV, whereas both pairwise comparison and slidingwindow analysis failed. This comparison was later found to
lack power in some genes because M1 does not account for
sites with 0 , v , 1 and the third class in M2 is forced to
account for such sites37. Thus, Yang et al.37 implemented
several new models. For example, the beta distribution
(M7 beta) is a flexible null model with 0 , v , 1, and can be
compared with an alternative that adds an additional site
class with v estimated (M8 beta&v). A general discrete
model (M3) was also implemented37. These models identified positive selection in six out of ten genes the authors
analysed. Figure 1 shows the use of a discrete model (M3)
with three classes to identify sites under diversifying selection in abalone sperm lysin39.
The methods discussed above assume that there are
heterogeneous classes of amino acid sites but that we do
not know a priori which class each site is from. Such fishing-expedition studies might be useful in generating
hypotheses for laboratory investigation because they
could identify crucial amino acids whose changes have
offered a selective advantage in Natures evolutionary
experiment. For example, amino acid residues under
diversifying selection were inferred in analyses of HIV-1 nef
(Ref. 38) and env (Ref. 40) genes, which might constitute
unidentified viral epitopes. Alternatively, we might wish to
test an a priori hypothesis that certain structural and functional domains of the protein are under positive selection.
In such cases, likelihood models can be constructed that
assign and estimate different v parameters for sites from
different structural and functional domains20.

Limitations of current methods and future directions

All the methods for detecting positive selection reviewed
here appear to be conservative. They detect selection only
if dN is higher than dS selection that does not cause excessive nonsynonymous substitutions, such as balancing
selection, might not be detected. The pairwise comparison
has little power because it averages the v ratio over sites
and over time. Methods for detecting selection along lineages work only if the v ratio averaged over all sites is .1.
Similarly, the test of positive selection at sites works only if
the v ratio averaged over all branches is .1. If adaptive
evolution occurs only in a short time interval and affects
only a few crucial amino acids, none of the methods is
likely to succeed. Constancy of selective pressure at sites
appears to be a much more serious assumption than constancy among lineages, especially for genes likely to be
under continuous selective pressure, such as the HIV env
gene. Indeed, models of variable selective pressures
among sites36,37 have been successful in detecting positive
selection, even in a background of overwhelming purifying
selection indicated by an average v ratio much smaller
than one37,38,41,42. Models that allow v to vary among both
lineages and sites should have increased power.
The methods discussed here also assume the same
v ratio for all possible amino acid changes; for example,
at a positively selected site, all amino acid changes
are assumed to be advantageous, which is unrealistic.
Although amino acid substitution rates are known to correlate with their chemical properties, the relationship is
poorly understood43,44. It is also not entirely clear how to
define positive selection in a model accounting for
chemical properties.
It will be interesting to perform computer simulations to
examine the power of various detection methods and to
TREE vol. 15, no. 12 December 2000

investigate how this is affected by important factors, such as

the size of the gene, sampling of species (sequences) and
the level of sequence divergence. Including more sequences
in the data should improve the power of site-based analyses.
Sequence divergence is also important because neither very
similar nor very divergent sequences contain much information. Very divergent sequences might also be associated
with problems with alignment and unequal nucleotide compositions in different species. Analyses discussed here,
which require information from both synonymous and nonsynonymous substitutions, are expected to have a narrower
window of suitable sequence divergences than phylogeny
reconstruction. The large-sample x2 approximation to the
likelihood-ratio test statistic might also be examined, but
limited simulations suggest that typical sequence data (with
.100 codons) are large enough for it to be reliable. For very
short genes or gene regions and especially at low sequence
divergences, Monte Carlo simulation might be needed to
derive the null distribution.
The likelihood analysis assumes no recombination
within a gene. If recombination occurs, different regions will
have different phylogenies. Empirical data analysis suggests
that the phylogeny does not have much impact on tests of
positive selection and identification of sites, and one might
suspect that recombination will not cause false positives by
the likelihood-ratio test. However, simulation studies are
necessary to understand whether this is the case.
Acknowledgements
We thank D. Haydon, J. Mallet, T. Ohta, A. Pomiankowski,
V. Vacquier, W. Swanson and three anonymous referees for
comments. We also thank several users of the PAML package
(http://abacus.gene.ucl.ac.uk/software/paml.html), in particular
C. Woelk, for comments and suggestions concerning the
implementation. This work is supported by grant #31/G10434
from the Biotechnology and Biological Sciences Research
Council (UK).
References
1 Lewontin, R.C. (1979) Adaptation. Sci. Am. 239, 156169
2 Kimura, M. (1983) The Neutral Theory of Molecular Evolution,
Cambridge University Press
3 Endo, T. et al. (1996) Large-scale search for genes on which positive
selection may operate. Mol. Biol. Evol. 13, 685690
4 Hughes, A.L. (1999) Adaptive Evolution of Genes and Genomes,
Oxford University Press
5 Miyata, T. and Yasunaga, T. (1980) Molecular evolution of mRNA: a
method for estimating evolutionary rates of synonymous and amino
acid substitutions from homologous nucleotide sequences and its
applications. J. Mol. Evol. 16, 2336
6 Akashi, H. (1995) Inferring weak selection from patterns of
polymorphism and divergence at silent sites in Drosophila DNA.
Genetics 139, 10671076
7 Kreitman, M. and Akashi, H. (1995) Molecular evidence for natural
selection. Annu. Rev. Ecol. Syst. 26, 403422
8 Wayne, M.L. and Simonsen, K.L. (1998) Statistical tests of neutrality in
the age of weak selection. Trends Ecol. Evol. 13, 236240
9 Nei, M. and Gojobori, T. (1986) Simple methods for estimating the
numbers of synonymous and nonsynonymous nucleotide
substitutions. Mol. Biol. Evol. 3, 418426
10 Li, W-H. et al. (1985) A new method for estimating synonymous and
nonsynonymous rates of nucleotide substitutions considering the
relative likelihood of nucleotide and codon changes. Mol. Biol. Evol. 2,
150174
11 Li, W-H. (1993) Unbiased estimation of the rates of synonymous and
nonsynonymous substitution. J. Mol. Evol. 36, 9699
12 Pamilo, P. and Bianchi, N.O. (1993) Evolution of the Zfx and Zfy genes
rates and interdependence between the genes. Mol. Biol. Evol. 10,
271281

501

REVIEWS
13 Comeron, J.M. (1995) A method for estimating the numbers of
synonymous and nonsynonymous substitutions per site. J. Mol. Evol.
41, 11521159
14 Ina, Y. (1995) 0New methods for estimating the numbers of
synonymous and nonsynonymous substitutions. J. Mol. Evol. 40,
190226
15 Yang, Z. and Nielsen, R. (2000) Estimating synonymous and
nonsynonymous substitution rates under realistic evolutionary
models. Mol. Biol. Evol. 17, 3243
16 Goldman, N. and Yang, Z. (1994) A codon-based model of nucleotide
substitution for protein-coding DNA sequences. Mol. Biol. Evol. 11,
725736
17 Moriyama, E.N. and Powell, J.R. (1997) Synonymous substitution rates
in Drosophila: mitochondrial versus nuclear genes. J. Mol. Evol. 45,
378391
18 Yang, Z. and Nielsen, R. (1998) Synonymous and nonsynonymous rate
variation in nuclear genes of mammals. J. Mol. Evol. 46, 409418
19 Muse, S.V. (1996) Estimating synonymous and nonsynonymous
substitution rates. Mol. Biol. Evol. 13, 105114
20 Yang, Z. (2000) Adaptive molecular evolution. In Handbook of
Statistical Genetics (Balding, D. et al., eds), Ch. 12, Wiley
21 Bielawski, J. et al. Rates of nucleotide substitution and mammalian
nuclear gene evolution: approximate and maximum-likelihood
methods lead to different conclusions. Genetics (in press)
22 Gillespie, J.H. (1991) The Causes of Molecular Evolution, Oxford
University Press
23 Messier, W. and Stewart, C-B. (1997) Episodic adaptive evolution of
primate lysozymes. Nature 385, 151154
24 Yang, Z. et al. (1995) A new method of inference of ancestral
nucleotide and amino acid sequences. Genetics 141, 16411650
25 Koshi, J.M. and Goldstein, R.A. (1996) Probabilistic reconstruction of
ancestral protein sequences. J. Mol. Evol. 42, 313320
26 Crandall, K.A. and Hillis, D.M. (1997) Rhodopsin evolution in the dark.
Nature 387, 667668
27 Yang, Z. (1998) Likelihood ratio tests for detecting positive selection
and application to primate lysozyme evolution. Mol. Biol. Evol. 15,
568573
28 McDonald, J.H. and Kreitman, M. (1991) Adaptive protein evolution at
the Adh locus in Drosophila. Nature 351, 652654
29 Hasegawa, M. et al. (1998) Preponderance of slightly deleterious
polymorphism in mitochondrial DNA: replacement/synonymous rate
ratio is much higher within species than between species. Mol. Biol.
Evol. 15, 14991505
30 Ohta, T. (1973) Slightly deleterious mutant substitutions in evolution.
Nature 246, 9698
31 Golding, G.B. and Dean, A.M. (1998) The structural basis of molecular
adaptation. Mol. Biol. Evol. 15, 355369
32 Chang, B.S. and Donoghue, M.J. (2000) Recreating ancestral proteins.
Trends Ecol. Evol. 15, 109114
33 Fitch, W.M. et al. (1997) Long term trends in the evolution of H(3) HA1
human influenza type A. Proc. Natl. Acad. Sci. U. S. A. 94, 77127718
34 Bush, R.M. et al. (1999) Positive selection on the H3 hemagglutinin
gene of human influenza virus A. Mol. Biol. Evol. 16, 14571465
35 Suzuki, Y. and Gojobori, T. (1999) A method for detecting positive
selection at single amino acid sites. Mol. Biol. Evol. 16, 13151328
36 Nielsen, R. and Yang, Z. (1998) Likelihood models for detecting
positively selected amino acid sites and applications to the HIV-1
envelope gene. Genetics 148, 929936
37 Yang, Z. et al. (2000) Codon-substitution models for heterogeneous
selection pressure at amino acid sites. Genetics 155, 431449
38 Zanotto, P.M. et al. (1999) Genealogical evidence for positive selection
in the nef gene of HIV-1. Genetics 153, 10771089
39 Yang, Z. et al. (2000) Maximum likelihood analysis of molecular
adaptation in abalone sperm lysin reveals variable selective pressures
among lineages and sites. Mol. Biol. Evol. 17, 14461455
40 Yamaguchi-Kabata, Y. and Gojobori, T. (2000) Reevaluation of amino
acid variability of the human immunodeficiency virus type 1 gp120
envelope glycoprotein and prediction of new discontinuous epitopes.
J. Virol. 74, 43354350
41 Bishop, J.G. et al. (2000) Rapid evolution in plant chitinases: molecular
targets of selection in plantpathogen coevolution. Proc. Natl. Acad.
Sci. U. S. A. 97, 53225327
42 Haydon, D.T. et al. Evidence for positive selection in foot-and-mouthdisease virus capsid genes from field isolates. Genetics (in press)
43 Yang, Z. et al. (1998) Models of amino acid substitution and applications
to mitochondrial protein evolution. Mol. Biol. Evol. 15, 16001611

502

44 Zhang, J. (2000) Rates of conservative and radical nonsynonymous

nucleotide substitutions in mammalian nuclear genes. J. Mol. Evol. 50, 5668
45 Riley, M.A. (1993) Positive selection for colicin diversity in bacteria.
Mol. Biol. Evol. 10, 10481059
46 Hughes, A.L. and Yeager, M. (1997) Coordinated amino acid changes
in the evolution of mammalian defensins. J. Mol. Evol. 44, 675682
47 Qi, C.F. et al. (1998) Molecular phylogeney of Fv1. Mamm. Genome 9,
10491055
48 Tanaka, T. and Nei, M. (1989) Positive darwinian selection observed at the
variable-region genes of immunoglobulins. Mol. Biol. Evol. 6, 447459
49 Hughes, A.L. and Nei, M. (1988) Pattern of nucleotide substitution at
major histocompatibility complex class I loci reveals overdominant
selection. Nature 335, 167170
50 Stotz, H.U. et al. (2000) Identification of target amino acids that affect
interactions of fungal polygalacturonases and their plant inhibitors.
Mol. Physiol. Plant Path. 56, 117130
51 Kitano, T. et al. (1998) Conserved evolution of the Rh50 gene
compared to its homologous Rh blood group gene. Biochem. Biophys.
Res. Commun. 249, 7885
52 Zhang, J. et al. (1998) Positive darwinian selection after gene
duplication in primate ribonuclease genes. Proc. Natl. Acad. Sci.
U. S. A. 95, 37083713
53 Ford, M.J. et al. (1999) Natural selection promotes divergence of
transferrin among salmonid species. Mol. Ecol. 8, 10551061
54 Hughes, A.L. (1995) The evolution of the type I interferon family in
mammals. J. Mol. Evol. 41, 539548
55 Goodwin, R.L. et al. (1996) Patterns of divergence during evolution of
a1-proteinase inhibitors in mammals. Mol. Biol. Evol. 13, 346358
56 Hughes, M.K. and Hughes, A.L. (1995) Natural selection on
Plasmodium surface proteins. Mol. Biochem. Parasitol. 71, 99113
57 Wu, J.C. et al. (1999) Recombination of hepatitis D virus RNA
sequences and its implications. Mol. Biol. Evol. 16, 16221632
58 Hughes, A.L. (1992) Positive selection and interallelic recombination at
the merozoite surface antigen-1 (MSA-1) locus of Plasmodium
falciparum. Mol. Biol. Evol. 9, 381393
59 Smith, N.H. et al. (1995) Sequence evolution of the porB gene of
Neisseria gonorrhoeae and Neisseria meningitidis: evidence for
positive darwinian selection. Mol. Biol. Evol. 12, 363370
60 Baric, R.S. et al. (1997) Episodic evolution mediates interspecific
transfer of a murine coronavirus. J. Virol. 71, 19461955
61 Vacquier, V.D. et al. (1997) Positive darwinian selection on two
homologous fertilization proteins: what is the selective pressure
driving their divergence? J. Mol. Evol. 44, 1522
62 Tsaur, S.C. and Wu, C-I. (1997) Positive selection and the molecular
evolution of a gene of male reproduction, Acp26Aa of Drosophila. Mol.
Biol. Evol. 14, 544549
63 Karn, R.C. and Nachman, M.W. (1999) Reduced nucleotide variability at
an androgen-binding protein locus (Abpa) in house mice: evidence for
positive natural selection. Mol. Biol. Evol. 16, 11921197
64 Metz, E.C. and Palumbi, S.R. (1996) Positive selection and sequence
arrangements generate extensive polymorphism in the gamete
recognition protein bindin. Mol. Biol. Evol. 13, 397406
65 Ting, C.T. et al. (1998) A rapidly evolving homeobox at the site of a
hybrid sterility gene. Science 282, 15011504
66 Sutton, K.A. and Wilkinson, M.F. (1997) Rapid evolution of a
homeodomain: evidence for positive selection. J. Mol. Evol. 45, 579588
67 Rooney, A.P. and Zhang, J. (1999) Rapid evolution of a primate sperm
protein: relaxation of functional constraint or positive darwinian
selection? Mol. Biol. Evol. 16, 706710
68 Ishimizu, T. et al. (1998) Identification of regions in which positive
selection may operate in S-RNase of Rosaceae: implications for Sallele-specific recognition sites in S-RNase. FEBS Lett. 440, 337342
69 Pamilo, P. and ONeill, R.W. (1997) Evolution of Sry genes. Mol. Biol.
Evol. 14, 4955
70 Ward, T.J. et al. (1997) Nucleotide sequence evolution at the k-casein
locus: evidence for positive selection within the family Bovidae.
Genetics 147, 18631872
71 Duda, T.F., Jr and Palumbi, S.R. (1999) Molecular genetics of ecological
diversification: duplication and rapid evolution of toxin genes of the
venomous gastropod Conus. Proc. Natl. Acad. Sci. U. S. A. 96, 68206823
72 Nakashima, K. et al. (1995) Accelerated evolution in the protein-coding
regions is universal in crotalinae snake venom gland phospholipase A2
isozyme genes. Proc. Natl. Acad. Sci. U. S. A. 92, 56055609

TREE vol. 15, no. 12 December 2000

REVIEWS
73 Schmidt, T.R. et al. (1999) Molecular evolution of the COX7A gene
family in primates. Mol. Biol. Evol. 16, 619626
74 Wu, W. et al. (1997) Molecular evolution of cytochrome c oxidase
subunit IV: evidence for positive selection in simian primates. J. Mol.
Evol. 44, 477491
75 Shields, D.C. et al. (1996) Evolution of hemopoietic ligands and their
receptors: influence of positive selection on correlated replacements
throughout ligand and receptor proteins. J. Immunol. 156, 10621070
76 Wallis, M. (1996) The molecular evolution of vertebrate growth
hormones: a pattern of near-stasis interrupted by sustained bursts of

rapid changes. J. Mol. Evol. 43, 93100

77 Bargelloni, L. et al. (1998) Antarctic fish hemoglobins: evidence for
adaptive evolution at subzero temperatures. Proc. Natl. Acad. Sci.
U. S. A. 95, 86708675
78 Long, M. and Langley, C.H. (1993) Natural selection and the origin of
jingwei, a chimeric processed functional gene in Drosophila. Science
260, 9195
79 Muse, S.V. and Gaut, B.S. (1994) A likelihood approach for comparing
synonymous and nonsynonymous nucleotide substitution rates, with
application to the chloroplast genome. Mol. Biol. Evol. 11, 715724

Nice snake, shame about the legs

Michael Coates and Marcello Ruta

relative to the retina. Moreover,

unlike lizards, snakes lack both a
fovea and coloured oil droplets in
retinal cells1.
Alternative hypotheses5 postulate that snakes are related
to mosasauroids (Fig. 1c): spectacular marine reptiles from the
upper half of the Cretaceous
period, some 65100 Mya6.
Mosasauroids and snakes share
reduced ossification of the pelvis
and hindlimbs as well as specialized features of the jaw suspension and intramandibular joint
kinetics (presence of a hinge
allowing a degree of lateral
movement within the lower jaw;
Fig. 1a,c,d; Fig. 2, red circle).
Phylogenetically, mosasauroids
would be the nearest monophyletic sister group of snakes,
with varanoid lizards (monitors)
Michael Coates and Marcello Ruta are at the Dept of
as the immediate sister group to
Biology, Darwin Building, University College London,
this pair. Given this theory of
Gower Street, London, UK WC1E 6BT
relationships, the latest common
(m.coates@ucl.ac.uk; m.ruta@ucl.ac.uk).
Ancestral diggers or swimmers?
ancestor of mosasaurs and
Hypotheses concerning snake
snakes has been argued to have
interrelationships fall into two
been a limbed, aquatic or semimain groups. For some researchers, snakes descend from aquatic squamate5,711. Note that the implied ecological
terrestrial squamates that developed fossorial (burrow- shift from an aquatic to a terrestrial environment in snake
ing) habits. Two groups of lizards exhibiting such habitats, ancestry suggests that mosasaurs (implied) aquatic
amphisbaenians and dibamids, have often been regarded habits were also primitive for Serpentes. Subsequently,
as snakes closest living relatives4. Amphisbaenians, in snakes reduced and lost their limbs, although rudiments
particular, resemble scaly, loose-skinned earthworms, of the posterior pair remain in some forms, such
whose shovel-shaped or wedge-like heads function as as pythons.
soil-shunting devices. Specializations shared by snakes
(Fig. 1a), amphisbaenians (Fig. 1b) and dibamids include Fossils: perfect missing links...
loss, reduction and consolidation of skull bones; brain- Renewed interest in the origin of snakes has been triggered
case enclosure; dorsal displacement of jaw-closing mus- by the recognition and discovery of three remarkable fossil
cles; loss or reduction of limbs and girdles; and increased forms with hind legs. Each of these ancient snakes is
uniformity along the vertebral column. Furthermore, around 97 My old and originates from lowermost Upper
differences between the eyes of lizards and snakes are Cretaceous sediments in the Middle East.
consistent with a model in which structures that were
Pachyrhachis problematicus, from Israel (Fig. 1df),
barely useful in a burrower underwent progressive reduc- rapidly assumed a central position in debates about snake
tion. Thus, whereas lizards, like humans, distort eye lens phylogeny6,7,12. It has miniature hindlimbs articulated with a
curvature to focus on objects, snakes lack ciliary muscles rudimentary pelvic girdle (Fig. 1e,f), but sadly, its feet are
and are compelled to move the entire lens back and forth missing. Currently described from only two specimens, it
he evolutionary origin of
snakes (or Serpentes) has
been discussed for over 130
years and their phylogenetic position within squamates is still
debated. Around 2700 snake species are alive today and these are
divided into three main groups13
(Box 1): tiny fossorial (burrowing)
scolecophidians
(blindsnakes);
anilioids (pipesnakes), which are
mostly semi-fossorial; and macrostomatans, which include more
familiar taxa, such as boas,
pythons, vipers and cobras. In
addition to the more obvious diagnostic characters of body elongation, limblessness and jaws that
can engulf surprisingly large prey,
other key features of snakes
include absence of eyelids and
external ears, and the presence of
deeply forked tongues (linked to
their highly attuned and sophisticated chemosensory systems3).

TREE vol. 15, no. 12 December 2000

Snakes are one of the most extraordinary

groups of terrestrial vertebrates, with
numerous specializations distinguishing
them from other squamates (lizards and
their allies). Their musculoskeletal system
allows creeping, burrowing, swimming
and even gliding, and their predatory
habits are aided by chemo- and
thermoreceptors, an extraordinary degree
of cranial kinesis and, sometimes,
powerful venoms. Recent discoveries of
indisputable early fossil snakes with
posterior legs are generating intense
debate about the evolutionary origin of
these reptiles. New cladistic analyses
dispute the precise significance and
phylogenetic placement of these fossils.
These conflicting hypotheses imply
radically different scenarios of snake
origins and relationships with wide
biological implications.

0169-5347/00/$ see front matter 2000 Elsevier Science Ltd. All rights reserved.

PII: S0169-5347(00)01999-6

503